1 /*- 2 * Copyright (c) 2011 NetApp, Inc. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/linker_set.h> 34 #include <sys/stat.h> 35 #include <sys/uio.h> 36 #include <sys/ioctl.h> 37 #include <sys/disk.h> 38 39 #include <errno.h> 40 #include <fcntl.h> 41 #include <stdio.h> 42 #include <stdlib.h> 43 #include <stdint.h> 44 #include <string.h> 45 #include <strings.h> 46 #include <unistd.h> 47 #include <assert.h> 48 #include <pthread.h> 49 #include <md5.h> 50 51 #include "bhyverun.h" 52 #include "pci_emul.h" 53 #include "virtio.h" 54 55 #define VTBLK_RINGSZ 64 56 57 #define VTBLK_MAXSEGS 32 58 59 #define VTBLK_S_OK 0 60 #define VTBLK_S_IOERR 1 61 #define VTBLK_S_UNSUPP 2 62 63 #define VTBLK_BLK_ID_BYTES 20 64 65 /* Capability bits */ 66 #define VTBLK_F_SEG_MAX (1 << 2) /* Maximum request segments */ 67 #define VTBLK_F_BLK_SIZE (1 << 6) /* cfg block size valid */ 68 69 /* 70 * Host capabilities 71 */ 72 #define VTBLK_S_HOSTCAPS \ 73 ( VTBLK_F_SEG_MAX | \ 74 VTBLK_F_BLK_SIZE | \ 75 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */ 76 77 /* 78 * Config space "registers" 79 */ 80 struct vtblk_config { 81 uint64_t vbc_capacity; 82 uint32_t vbc_size_max; 83 uint32_t vbc_seg_max; 84 uint16_t vbc_geom_c; 85 uint8_t vbc_geom_h; 86 uint8_t vbc_geom_s; 87 uint32_t vbc_blk_size; 88 uint32_t vbc_sectors_max; 89 } __packed; 90 91 /* 92 * Fixed-size block header 93 */ 94 struct virtio_blk_hdr { 95 #define VBH_OP_READ 0 96 #define VBH_OP_WRITE 1 97 #define VBH_OP_IDENT 8 98 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */ 99 uint32_t vbh_type; 100 uint32_t vbh_ioprio; 101 uint64_t vbh_sector; 102 } __packed; 103 104 /* 105 * Debug printf 106 */ 107 static int pci_vtblk_debug; 108 #define DPRINTF(params) if (pci_vtblk_debug) printf params 109 #define WPRINTF(params) printf params 110 111 /* 112 * Per-device softc 113 */ 114 struct pci_vtblk_softc { 115 struct virtio_softc vbsc_vs; 116 pthread_mutex_t vsc_mtx; 117 struct vqueue_info vbsc_vq; 118 int vbsc_fd; 119 struct vtblk_config vbsc_cfg; 120 char vbsc_ident[VTBLK_BLK_ID_BYTES]; 121 }; 122 123 static void pci_vtblk_reset(void *); 124 static void pci_vtblk_notify(void *, struct vqueue_info *); 125 static int pci_vtblk_cfgread(void *, int, int, uint32_t *); 126 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t); 127 128 static struct virtio_consts vtblk_vi_consts = { 129 "vtblk", /* our name */ 130 1, /* we support 1 virtqueue */ 131 sizeof(struct vtblk_config), /* config reg size */ 132 pci_vtblk_reset, /* reset */ 133 pci_vtblk_notify, /* device-wide qnotify */ 134 pci_vtblk_cfgread, /* read PCI config */ 135 pci_vtblk_cfgwrite, /* write PCI config */ 136 NULL, /* apply negotiated features */ 137 VTBLK_S_HOSTCAPS, /* our capabilities */ 138 }; 139 140 static void 141 pci_vtblk_reset(void *vsc) 142 { 143 struct pci_vtblk_softc *sc = vsc; 144 145 DPRINTF(("vtblk: device reset requested !\n")); 146 vi_reset_dev(&sc->vbsc_vs); 147 } 148 149 static void 150 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) 151 { 152 struct virtio_blk_hdr *vbh; 153 uint8_t *status; 154 int i, n; 155 int err; 156 int iolen; 157 int writeop, type; 158 off_t offset; 159 struct iovec iov[VTBLK_MAXSEGS + 2]; 160 uint16_t flags[VTBLK_MAXSEGS + 2]; 161 162 n = vq_getchain(vq, iov, VTBLK_MAXSEGS + 2, flags); 163 164 /* 165 * The first descriptor will be the read-only fixed header, 166 * and the last is for status (hence +2 above and below). 167 * The remaining iov's are the actual data I/O vectors. 168 * 169 * XXX - note - this fails on crash dump, which does a 170 * VIRTIO_BLK_T_FLUSH with a zero transfer length 171 */ 172 assert(n >= 2 && n <= VTBLK_MAXSEGS + 2); 173 174 assert((flags[0] & VRING_DESC_F_WRITE) == 0); 175 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr)); 176 vbh = iov[0].iov_base; 177 178 status = iov[--n].iov_base; 179 assert(iov[n].iov_len == 1); 180 assert(flags[n] & VRING_DESC_F_WRITE); 181 182 /* 183 * XXX 184 * The guest should not be setting the BARRIER flag because 185 * we don't advertise the capability. 186 */ 187 type = vbh->vbh_type & ~VBH_FLAG_BARRIER; 188 writeop = (type == VBH_OP_WRITE); 189 190 offset = vbh->vbh_sector * DEV_BSIZE; 191 192 iolen = 0; 193 for (i = 1; i < n; i++) { 194 /* 195 * - write op implies read-only descriptor, 196 * - read/ident op implies write-only descriptor, 197 * therefore test the inverse of the descriptor bit 198 * to the op. 199 */ 200 assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop); 201 iolen += iov[i].iov_len; 202 } 203 204 DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r", 205 writeop ? "write" : "read/ident", iolen, i - 1, offset)); 206 207 switch (type) { 208 case VBH_OP_WRITE: 209 err = pwritev(sc->vbsc_fd, iov + 1, i - 1, offset); 210 break; 211 case VBH_OP_READ: 212 err = preadv(sc->vbsc_fd, iov + 1, i - 1, offset); 213 break; 214 case VBH_OP_IDENT: 215 /* Assume a single buffer */ 216 strlcpy(iov[1].iov_base, sc->vbsc_ident, 217 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident))); 218 err = 0; 219 break; 220 default: 221 err = -ENOSYS; 222 break; 223 } 224 225 /* convert errno into a virtio block error return */ 226 if (err < 0) { 227 if (err == -ENOSYS) 228 *status = VTBLK_S_UNSUPP; 229 else 230 *status = VTBLK_S_IOERR; 231 } else 232 *status = VTBLK_S_OK; 233 234 /* 235 * Return the descriptor back to the host. 236 * We wrote 1 byte (our status) to host. 237 */ 238 vq_relchain(vq, 1); 239 } 240 241 static void 242 pci_vtblk_notify(void *vsc, struct vqueue_info *vq) 243 { 244 struct pci_vtblk_softc *sc = vsc; 245 246 vq_startchains(vq); 247 while (vq_has_descs(vq)) 248 pci_vtblk_proc(sc, vq); 249 vq_endchains(vq, 1); /* Generate interrupt if appropriate. */ 250 } 251 252 static int 253 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts) 254 { 255 struct stat sbuf; 256 MD5_CTX mdctx; 257 u_char digest[16]; 258 struct pci_vtblk_softc *sc; 259 off_t size; 260 int fd; 261 int sectsz; 262 263 if (opts == NULL) { 264 printf("virtio-block: backing device required\n"); 265 return (1); 266 } 267 268 /* 269 * The supplied backing file has to exist 270 */ 271 fd = open(opts, O_RDWR); 272 if (fd < 0) { 273 perror("Could not open backing file"); 274 return (1); 275 } 276 277 if (fstat(fd, &sbuf) < 0) { 278 perror("Could not stat backing file"); 279 close(fd); 280 return (1); 281 } 282 283 /* 284 * Deal with raw devices 285 */ 286 size = sbuf.st_size; 287 sectsz = DEV_BSIZE; 288 if (S_ISCHR(sbuf.st_mode)) { 289 if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 || 290 ioctl(fd, DIOCGSECTORSIZE, §sz)) { 291 perror("Could not fetch dev blk/sector size"); 292 close(fd); 293 return (1); 294 } 295 assert(size != 0); 296 assert(sectsz != 0); 297 } 298 299 sc = calloc(1, sizeof(struct pci_vtblk_softc)); 300 301 /* record fd of storage device/file */ 302 sc->vbsc_fd = fd; 303 304 pthread_mutex_init(&sc->vsc_mtx, NULL); 305 306 /* init virtio softc and virtqueues */ 307 vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq); 308 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx; 309 310 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ; 311 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */ 312 313 /* 314 * Create an identifier for the backing file. Use parts of the 315 * md5 sum of the filename 316 */ 317 MD5Init(&mdctx); 318 MD5Update(&mdctx, opts, strlen(opts)); 319 MD5Final(digest, &mdctx); 320 sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X", 321 digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]); 322 323 /* setup virtio block config space */ 324 sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */ 325 sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS; 326 sc->vbsc_cfg.vbc_blk_size = sectsz; 327 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */ 328 sc->vbsc_cfg.vbc_geom_c = 0; /* no geometry */ 329 sc->vbsc_cfg.vbc_geom_h = 0; 330 sc->vbsc_cfg.vbc_geom_s = 0; 331 sc->vbsc_cfg.vbc_sectors_max = 0; 332 333 /* 334 * Should we move some of this into virtio.c? Could 335 * have the device, class, and subdev_0 as fields in 336 * the virtio constants structure. 337 */ 338 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK); 339 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); 340 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE); 341 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK); 342 343 pci_lintr_request(pi); 344 345 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) 346 return (1); 347 vi_set_io_bar(&sc->vbsc_vs, 0); 348 return (0); 349 } 350 351 static int 352 pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value) 353 { 354 355 DPRINTF(("vtblk: write to readonly reg %d\n\r", offset)); 356 return (1); 357 } 358 359 static int 360 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval) 361 { 362 struct pci_vtblk_softc *sc = vsc; 363 void *ptr; 364 365 /* our caller has already verified offset and size */ 366 ptr = (uint8_t *)&sc->vbsc_cfg + offset; 367 memcpy(retval, ptr, size); 368 return (0); 369 } 370 371 struct pci_devemu pci_de_vblk = { 372 .pe_emu = "virtio-blk", 373 .pe_init = pci_vtblk_init, 374 .pe_barwrite = vi_pci_write, 375 .pe_barread = vi_pci_read 376 }; 377 PCI_EMUL_SET(pci_de_vblk); 378